1

Nature as Laboratory: Experiments in Ecology and Evolutionary Biology

Bradley E. WilsonDepartment of PhilosophySlippery Rock University

Presented at the Second Biennial Conference of the Society for the Philosophy of Science in Practice, June 18-20, 2009, Minneapolis, Minnesota.

“I was talking with someone in vision physiology the other day and he told me, ‘Wow, I had no idea that the subject [evolutionary biology] was so rigorous. I had no idea that you actually did experiments.” (John Endler, quoted in Jonathan Weiner, The Beak of the Finch)

I. Introduction

Experimentation has been a central part of scientific practice since at least the beginning

of the Royal Society of London. However, until recently philosophers of science have paid

relatively little attention to experiments and experimentation as fit topics for philosophical

discussion in comparison to topics such as the structure of scientific theories, scientific

explanation, or the logic of confirmation. This lack of attention to experimentation is puzzling,

given the central role that experiments play in the actual practice of science.

Two reasons for the lack of attention come to mind. First, it may be that philosophers

have not found any interesting philosophical problems associated with experiments that are not

equally well-addressed in other contexts. So, for example, it may be that experimentation is

understood to be a (potentially complex) form of observation, and thus the philosophical

problems associated with experimentation are best addressed by focusing on issues related to the

distinction between theory and observation. The second, and I think more important, reason for

2

the lack of attention to experimentation has to do with the remarkable diversity of the kinds of

experiments performed in different fields and subfields of science. In the size and complexity of

the experimental setups, physical and chemical experiments range from those that can be done

on a bench top to experiments in high-energy physics that use enormous particle accelerators. If

we add to these experiments in the life sciences and the social sciences, the possibility of

developing a general philosophy of experiment appears quite daunting.

Fortunately, the situation is improving. Beginning with the work of Ian Hacking (1983),

there has been a steady increase in interest in the role of experiment in science and its attendant

philosophical problems. Much of the work that has been done on experimentation has focused

on the physical and chemical sciences, so I will begin by sketching a view of experiments that is

based on some of that work. However, my main interest is in thinking about the extent to which

a physics-based understanding of experimentation can be helpfully applied to experiments in

ecology and evolutionary biology. While there are undoubtedly points of overlap in our

understanding of experimentation throughout the sciences, I hope to elucidate some significant

underlying differences between the physical sciences and ecology and evolutionary biology

(hereafter E&EB) in terms of the role of experiments in these sciences. Not surprisingly, the

main differences stem from the fact that many of the experiments done in E&EB are done in the

field, rather than the laboratory. There are both advantages and disadvantages to field

experiments. On the one hand, in field experiments the connection between the experimental

system and the natural system being studied is very close. Thus, results gained from field

experiments apply quite directly to actual systems in nature. However, the downside is that it is

difficult to generalize to the results of field experiments to systems other than those that are

being directly studied. This has important implications for our understanding of the role of

general theories in ecology and evolutionary biology.

3

II. An experimental paradigm: Millikan’s Oil-Drop Experiment

As mentioned above, much of the work by philosophers on experimentation has focused

on the physical and chemical sciences. Robert Millikan’s Oil-Drop experiment provides a good

example of an experiment in physics. In 1911, Millikan published the results of an experiment

that he had conducted (with a graduate student, --- Fletcher) to measure the charge of an electron.

Previous attempts to measure this charge had used a cloud chamber, and had been limited in their

accuracy by the necessity of measuring the charge on a large number of water droplets and then

averaging to find the value for a single ion. Millikan improved this method by developing a

technique by which he could measure the velocity of a falling charged water droplet in an

electric field, from which he could determine the magnitude of the charge. However, water

droplets evaporate relatively quickly, making accurate calculations difficult. So Millikan

modified the technique again by using droplets of oil rather than water, which could be

suspended between the electric plates of his apparatus for indefinite periods of time.

Millikan’s experiment exemplifies many of the characteristics of experiments in the

physical sciences. First, his experiments took place in a laboratory using a sophisticated

apparatus developed for the purpose of studying the behavior of ionized substances.

Furthermore, he created in the laboratory, within his apparatus, very special conditions that

would enable him to isolate the phenomena that he was studying. In the course of several hours

of viewing, each oil droplet could capture and release ions, which would be observable in terms

of movement within the electrical field. Thirdly, his experiment was not created de novo, but

arose from the perceived importance within the scientific community of determining the charge

of the recently identified subatomic particle. In this particular case, it was not so much a test of a

hypothesis as it was the measurement of an important physical quantity. Focusing on these

4

elements of Millikan’s experiment will be helpful in identifying some of the main ideas that have

been proposed for a philosophical understanding of experimentation.

III. Three Kinds of Experiments in Ecology and Evolutionary Biology

Experiments in physics and chemistry are typically conducted in laboratories. Millikan’s

experiment was conducted in a laboratory at the University of Chicago and involved a

sophisticated apparatus to measure the velocity of charged oil drops. Millikan set up conditions

that had probably not been previously realized outside of the laboratory: it is unlikely that there

were many charged oil droplets suspended motionless by electrostatic charges outside of

Millikan’s apparatus. While some ecological experiments are conducted in laboratories (often

referred to as “bottle experiments”), many are conducted in the field. We need to examine more

carefully the types of experiments that ecologists perform.

Robert Brandon (1996) has developed a classification of studies in evolutionary biology

that can also be applied to experiments in ecology. He distinguishes two dimensions among

which studies can be placed. The first dimension is the degree of manipulation involved in the

study. Studies range from those that involve a high degree of manipulation, similar to laboratory

experiments in the physical sciences, to those that involve no manipulation, the so-called

“natural experiments” of the life sciences. The second dimension is based on whether or not the

study is used to test a hypothesis, as opposed to measure a parameter, or simply to describe some

important aspect of nature. (Although the second dimension may be understood as a sharp

dichotomy, Brandon makes a case for viewing it as a continuum as well.) Given the two

dimensions, there are roughly four types of studies: non-manipulative, non-hypothesis testing;

non-manipulative hypothesis testing; manipulative, non-hypothesis testing; and manipulative,

5

hypothesis testing. While all four types of studies have value, it is the last type that is the most

“experimental,” and that forms the basis of any experimental science.

The vast majority of experiments in the physical sciences involve a high degree of

manipulation. Millikan’s Oil Drop experiment is one example. Furthermore, the distinction

between hypothesis testing and non-hypothesis testing has less significance in the physical

sciences. Again, Millikan’s Oil Drop experiment was not testing a hypothesis, but was designed

to determine an important value, the charge of an electron. As I will discuss below, the

difference between physics and ecology with respect to the value of non-hypothesis testing

experiments is related to an important difference between the two disciplines regarding the role

of general theory. Because some physical values are constant, experiments establishing these

values (or measuring them more accurately) play an important role in physics.

Experimental ecologists have recognized distinctions similar to those made by Brandon,

though not always explicitly. Along the manipulation gradient, ecologists distinguish between

laboratory experiments, field experiments, and natural experiments (see Diamond, 1986).

Laboratory experiments in ecology are frequently termed “bottle experiments” because they

often consist of populations of small organisms (e.g., bacteria, yeasts, Drosophila) raised and

maintained in bottles or test tubes in the laboratory (see Gause, 1933, for a classic example).

Such experiments are most similar to the laboratory experiments of the physical sciences.

Natural experiments do not involve any manipulation, at least no intentional manipulation by the

experimenter. In natural experiments, naturally occurring perturbations of ecological systems

are used to gain a better understanding of the system in question. Often, these perturbations

occur on a scale beyond what can be manipulated by an experimenter (e.g. forest fires, volcanic

eruptions, etc.) Field experiments fall between the two. Field experiments involve the

manipulation of some aspects of the system by the experimenter (e.g. setting up regular plots in

6

which different combinations of plants are grown), but many aspects of the system are

uncontrolled and uncontrollable (amount of sunlight, rainfall, soil composition, etc.). The degree

to which aspects of the system are subject to control by the experimenter are important for

experimental design and analysis (see Hairston), but here we need only be aware of the

difference between field experiments and those that are conducted in the laboratory.

It is less clear that experimental ecologists distinguish between experiments that test

hypotheses and those that do not. A related distinction is sometimes made in terms of the ability

of experiments to test predictions. Peter Morin, for example, expresses the distinction between

hypothesis testing and non-hypothesis testing in terms of the differences in the ability of different

types of experiments to test predictions (see Morin, 1998). Dunham and Beaupre also focus on

the ability to test predictions in their distinction between experiments that are mechanism-free or

mechanism-vague, and those that are mechanism specific (see Dunham and Beaupre, 1998). For

Dunham and Beaupre, the relevant hypotheses in ecology are those that specify the mechanism

or mechanisms producing a pattern of interest to the ecologist. Thus, mechanism-free or

mechanism-vague experiments are akin to those that don’t test a hypothesis.

IV. Experimental Systems and Natural Systems

One of the central themes of Hacking’s discussion of experimentation is the idea that

phenomena are created in the laboratory. Hacking has in mind as his paradigm the experiments

of high-energy physics, in which super-colliders are used to produce various subatomic particles.

If Hacking is correct about the creation of phenomena in the laboratory, then there is an

interesting and important question about the relation between the phenomena created in the

laboratory. This question can be expressed in two ways, one metaphysical, the other

epistemological: Do the phenomena created in the laboratory exist outside of the laboratory?

7

How can we generalize from what occurs in the laboratory to what occurs in nature? Focusing

on these two questions in relation to field experiments in ecology and evolutionary biology will

highlight some of the important differences between experiments in the physical sciences and

those in E&EB.

It may seem odd to question whether or not the physical phenomena created in the

laboratory exist outside the laboratory. After all, whether it is in or out of the lab, it is still a

physical phenomenon. However, the ontological interest of the question becomes clear when we

think about a genuine example. Long after the periodic table of elements had been developed

and accepted, there were still certain elements of the table that had not yet been detected (some

of the trans-uranic elements). Part of the problem is that these elements are highly unstable and

thus, if they exist at all, they don’t persist for very long. However, creative chemists have been

able to construct apparatuses by means of which they have been able to bring into existence

some of these theoretically predicted transuranic elements (citations would be helpful). These

elements have been created in the laboratory, and may not have ever existed outside of the

situations in which they were created. By means of such experiments, chemists have been able

to demonstrate that certain theoretically expected elements are realizable, but that doesn’t tell us

anything about whether such elements are actually realized in nature. There is a general point

here about the systems and entities studied by physical scientists in the laboratory: they may

stand at quite a distance from actual systems in nature.

When we look at the relationship between experiments in E&EB and systems in nature,

in some ways we find a similar situation, but there are also important differences. If we focus on

laboratory experiments, we can find examples that parallel to some degree the example just

discussed. In a series of classic experiments using flour beetles, Wade demonstrated the

possibility that group selection could have an impact on the traits of a laboratory population

8

(Wade, 1977). In doing so, Wade was able to realize experimentally a model of group selection

that theoretically could have an effect on a population. Although Wade had demonstrated the

possibility of group selection in the laboratory, it was still an open question whether or not the

conditions required for group selection to operate exist outside of the laboratory.

When we turn to field experiments, we find one of the important differences between

experiments in the physical sciences and experiments in E&EB. Here, almost by definition, the

distance between the experimental system and natural systems diminishes. While it is still a

mistake to equate the experimental system with a natural system, the ontological distance is

much narrower. This is because the field experiment makes use of a natural system as a part of

the experiment itself (along with the apparatuses used for manipulating the natural system). The

field experiments conducted by ecologists and evolutionary biologists have a much more direct

connection with actual systems of interest in nature than do the experiments conducted by

physicists and chemists in the laboratory. Considering an example from evolutionary biology

will help to illustrate these points.

V. An Experiment in Evolutionary Biology

I now turn to consider a set of experiments in evolutionary biology in the light of the

preceding discussion of experiments in the physical sciences. One goal is to determine to what

extent the work on experimentation that comes out of the study of physics and chemistry will

apply to experiments in ecology and evolutionary biology. Furthermore, identifying important

differences between experimentation in the physical sciences and experimentation in E&EB will

give us some insight into the different goals of physicists and biologists.

John Endler has studied extensively the guppy, Poecilia reticulata Peters (Endler, 1978,

1980). These guppies are native to mountain streams in northeastern Venezuela, Margarita,

9

Trinidad, and Tobago. Endler studied the color patterns of the guppies, which exhibit a great

deal of variation in natural populations. These color patterns “consist of a mosaic of spots or

patches varying in color, size, position, and reflectivity” and are expressed only in adult males

(Endler, 1980, 76). The color patterns of the guppies are important for two competing reasons.

First, insofar as they match the background of the stream (i.e. the gravel streambed), they

provide camouflage from predators. On the other hand, the more conspicuous the males are, the

better able they are to attract female mates. Thus, the color patterns exhibited in a given

population appear to represent a balance between two selection pressures.

Field studies of the color patterns of guppies in relation to predators supported this

hypothesis (Endler, 1978). In ponds where there were no predators or only the “relatively

innocuous” Rivulus hartii (Cyprinodontidae), the number of spots per fish, as well as degree of

brightness of the coloration, were much greater than in ponds where there were more dangerous

predators (Crenicichla alta (Cichlidae)). However, as it would be very difficult to track changes

through time in the natural populations, Endler devised an experiment to model the mountain

streams in a greenhouse at Princeton University. There he constructed ten ponds, which were

initially stocked with 200 fish per pond bred from numerous pairs of guppies taken randomly

from 18 different localities in Trinidad and Venezuela. The stocking method was designed to

create ponds with populations of guppies containing a high level of genetic variation.

Once the ten ponds were stocked with guppies, the two types of predators were placed in

the ponds, and a census of various aspects of the color patterns of the fish was made at two

different times: 5 months (generation 3-4) and 14 months (generation 9-10). The main

expectation of the experiment was that “the color patterns should converge on the color patterns

of natural populations of guppies living with the same predators” (Endler, 1980: 78). And this is

just what Endler found in the experimental ponds. With respect to the number of spots, spot

10

length, height, and total area, and spot color, the experimental populations converged on the

natural stream populations.

Endler also performed a related experiment in the field. A small stream (Aripo 1) was

identified in Trinidad that contained the innocuous predator R. hartii, but no guppies. A sample

of 200 guppies from a nearby stream where the guppies were exposed to the more dangerous

predator C. alta. were transferred to Aripo 1, with the expectation that the “color patterns of the

transferred population should change from the typical high predation patterns found with C. alta

to the more complex and colorful patterns found with R. hartii” (Endler, 79). Again, the

expectations were met: the color patterns changed to resemble those in streams with only the

predator R. hartii and ceased to resemble the parental population still living in the stream with C.

alta. Endler concluded that the results of both experiments were clear: “[t]he results are what

are expected if the color patterns represent a shifting balance between sexual selection and

predator avoidance” (Endler, 83).

VI. The Ontology of Experiment

How does this example fit into the view of experiments developed using the Millikan oil

drop experiment? We can identify two different types of experiment, one that could be

considered to involve a “laboratory” (in an expanded sense) and another that takes place outside

of the laboratory. Consider first the laboratory-based experiments, namely, Endler’s ponds. We

can understand these experiments to fit broadly the model of an experiment in physics. While

the boundaries of the “laboratory” are much less clear, we can without too much difficulty

understand the series of ponds to be the laboratory for these experiments. Similarly, Endler’s

experiments involve some apparatus, which functions to create the conditions in which the

phenomena to be studied can exist. Finally, the experiments are clearly examples of models of

11

systems outside of the laboratory. The explicit goal is to reproduce the environment outside of

the laboratory in which the organisms live so that the effects of the processes operative in that

environment can be studied. Thus, like many experiments in the physical sciences, Endler’s

pond experiments involve bringing “samples” (moths and guppies) from outside of the

laboratory into the laboratory to study them within a model of the system in which they are

found.

However, recall the issue raised above regarding the ontology of physical experiments.

In many instances, the experimental phenomena created in the laboratory do not exist outside of

the laboratory environment. While it may be possible to draw inferences from the laboratory to

what goes on “in nature,” the basis for those inferences needs to be clarified. In Endler’s

experiments, there is an explicit attempt to model as accurately as is feasible the natural system

being studied, rather than to isolate a single aspect of a natural system. Even minor aspects of

the experimental setup, such as the use of natural gravel vs. the use of commercially available

colored gravel, were considered in relation to the goals modeling a natural system and being able

to control the experimental setup.

When we move to field experiments, the separation between the “experimental” system

and the natural system practically disappears. In transferring fish from one stream to another

nearby stream, Endler intervened with or manipulated the natural system, but he left virtually

every other feature of that system intact and undisturbed. The experimental system and the

natural system are virtually identical. There are no difficulties, then, in drawing inferences about

the natural system on the basis of what happens in the experimental system. Field experiments

tell us fairly directly about what happens in the natural world.

VII. The Epistemology of Experiment

12

Making inferences beyond the experimental system primarily concerns what I am calling

the epistemology of experimentation: How does an experiment give us knowledge beyond the

experimental system with which we are dealing. It is here that we find important differences

between the physical sciences and the fields of ecology and evolutionary biology. One way of

framing the issue here is in terms of generalizing beyond the experimental system. What

justifies such generalization in the different scientific disciplines?

On the face of it, given the ontological issues raised earlier, physical scientists would

seem to have a hard time generalizing from experimental systems. If the experimental system

creates phenomena that do not exist in nature, then it is unclear how one could generalize to

nature. The solution to this difficulty lies in recognizing the role that theories play in the

physical sciences and in the relation between theory and experiment. Experiments in the

physical sciences are deeply connected to the development of theory. When Millikan conducted

his oil-drop experiments, the existence of the electron was not in question; what was unknown

was the charge on the electron. Furthermore, the particles that physicists studied were thought to

form a class of identical entities. One electron is as good as another when it comes to

investigating the physical properties of electrons. It is this assumption, that physicists deal with

classes of phenomena, that enables physicists to move beyond the particular instance in the

laboratory to acquire knowledge of the world outside of the laboratory. Since every electron is

the same, whether it is in the laboratory or outside the laboratory, there is no special difficulty

involved in generalizing beyond the laboratory.

The situation is different in ecology and evolutionary biology. There has been a great

deal of debate among philosophers about the best way to understand the structure of physical

theories and the extent to which an understanding of theory based on physics can be extended to

the biological sciences. Whatever theories are understood to be, the development of a general

13

theory has been far less successful in ecology than in the physical sciences. In evolutionary

biology, the Darwinian theory of evolution by natural selection provides a central framework,

but even here there are issues related to the possibility of generalizing on the basis of

experiments. In both ecology and evolutionary biology, we cannot assume that the phenomena

being studied either in the laboratory or in the field are instances of a general class of phenomena

about which we can readily generalize.

Let’s return to Endler’s guppy experiments. Endler’s laboratory experiments using the

ponds supports a hypothesis about the selection pressures affecting the size and color of spots on

the guppies. His field experiment provides further support that these same selection pressures

operate in the natural environment. But we have only gained knowledge about the particular

system being studied. It is not clear to what extent Endler’s results could be generalized to other

populations of the same species of guppy in different geographic areas, to other guppy species,

or beyond to other animal species. The framework of the Darwinian theory of natural selection

directs the researcher to investigate the selection pressures operative in any particular population,

but the systems studied will differ with respect to what the particular selection pressures are.

In order to generalize the results of laboratory and field experiments in ecology and

evolutionary biology, we need to find some basis analogous to the operative assumption in the

physical sciences that we are dealing with a class of similar phenomena, rather than a particular

phenomenon. One possibility is to focus on the underlying mechanisms of evolutionary or

ecological processes, and to identify systems in which the same mechanisms are operative. This

approach to community ecology is expressed well by Leibold and Tessier (1998):

The key component of the mechanistic approach is identification of a critical set

of causal relations responsible for a particular result. . . . The primary goal . . . is to

postulate clearly focused and operationally defined, albeit not comprehensive,

14

models. Experiments are then employed sequentially to test the important

assumptions and predictions from the model and separate studies conducted to

extrapolate their application to a broader, more natural context. . . . [A] mechanistic

approach retains the notion that the causal bases of patterns must themselves be

understood before a hypothesis can be accepted as an explanation and presumes that

an understanding of causal relations is an important element that allows extrapolation

of results beyond the set of previously studied conditions. (Leibold and Tessier,

1998, pp. 97-98.)

In ecology and evolutionary biology, knowledge about underlying mechanisms may play a role

similar to that of general theory in the physical sciences.

VIII. Conclusion

Experimentation plays a central role in all of science. Here I’ve tried to spell out the role

experiments, particularly those done in the field, play in ecology and evolutionary biology in a

way that highlights some important differences between the life sciences and the physical

sciences. More work needs to be done to understand both the role of general theory and the

understanding of mechanisms in E & EB before the role of experimentation can be fully

understood. It may be that sciences, such as ecology and evolutionary biology, that rely on

experiments that are not connected to more general theories are sciences that deal with “a

collection of special cases” (Dunham and Beaupre, 1998).

15

Works Cited

Brandon, R.N. 1996. Concepts and Methods in Evolutionary Biology. Cambridge: Cambridge

University Press.

Diamond, J. M. 1986. “Overview: laboratory experiments, field experiments, and natural

experiments,” in J. Diamond and T.J. Case (eds.), Community Ecology, New York:

Harper and Row, pp. 3-22.

Dunham A.E. and S.J. Beaupre. 1998. “Ecological experiments: scale, phenomenology,

mechanism, and the illusion of generality,” in Resetarits, Jr., W.J. and J. Bernardo (eds.),

Experimental Ecology: Issues and Perspectives. New York: Oxford, pp. 27-49.

Endler, J. 1978. “A Predator’s View of Animal Color Patterns,” Evolutionary Biology 11: 319-

364.

Endler, J. 1980. “Natural Selection on Color Patterns in Poecilia reticulata,” Evolution 34: 76-

91.

Gause, F. 1934. The Struggle for Existence. Baltimore: Williams and Wilkins.

Hacking, I. 1983. Representing and Intervening. Cambridge: Cambridge University Press.

Hairston, N. G. 1989. Ecological Experiments. Cambridge: Cambridge University Press.

Liebold, M.A. and A.J. Tessier. 1998. “Experimental Compromise and Mechanistic

Approaches to the Evolutionary Ecology of Interacting Daphnia species,” in Resetarits,

Jr., W.J. and J. Bernardo (eds.), Experimental Ecology: Issues and Perspectives. New

York: Oxford, pp. 96-112.

Morin, Peter J. 1998. “Realism, Precision, and Generality in Experimental Ecology,” in

Experimental Ecology: Issues and Perspectives, Resetarits, W.J. Jr. and Joseph Bernardo,

(eds.), New York: Oxford University Press, pp. 50-70.

Wade, M.J. 1977. “An Experimental Study of Group Selection,” Evolution 31: 134-153.